基于主曲线的脱机手写数字结构特征分析及选取-同济大学.PDF

计算机研究与发展 ISSN CN 111777/ TP Journal of Computer Research and Develop ment 42 (8) : 1344～1349 , 2005 基于主曲线的脱机手写数字结构特征分析及选取 张红云 　苗夺谦 　张东星 ( 同济大学计算机科学与技术系 　上海 　200092) (zhanghongyun583 @sinacom) Analysis and Extraction of Structural Features of OffL ine Handwritten Digits Based on Principal Curves Zhang Hongyun , Miao Duoqian , and Zhang Dongxing ( ) Dep art ment of Comp uter S cience an d Technology , Tongj i University , S hang hai 200092 Abstract 　Ext raction and choice of feat ures are critical to imp roving t he recognition rate of offline handwritten digit s Princip al curves are nonlinear generalizations of p rincip al component s analysis They are smoot h selfconsistent curves t hat p ass t hrough t he “middle ”of t he dist ribution They p referably reflect t he st ruct ural feat ures of t he dat a During digit feat ure selection , fir stly p rincip al curves are used to ext ract t he st ruct ural feat ures of t raining dat a ; Secondly t he classification feat ures used for digit s coar se classification and p recise classification are cho sen by analyzing t he st ruct ural feat ures of p rincip al curves in det ail ; Finally coar se classification and p recise classification are sep arately carried out in handwritten digit s recognition The Concor dia U niver sit y CEN PA RM I handwritten digit dat abase is used in t he experiment The result of t he experiment show s t hat t hese feat ures have good discriminating power of similar digit s The p ropo sed met hod can effectively imp rove t he recognition rate of offline handwritten digit s and p rovide a new app roach to t he research for offline handwritten digit s recognition